Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add tsubotax/melta-ui --skill build-screengit clone --depth 1 https://github.com/tsubotax/melta-uiWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/tsubotax/melta-ui/build-screen)<a href="https://agentmods.dev/skills/tsubotax/melta-ui/build-screen"><img src="https://agentmods.dev/badge/skills/tsubotax/melta-ui/build-screen/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/tsubotax/melta-ui/build-screen"><img src="https://agentmods.dev/badge/skills/tsubotax/melta-ui/build-screen.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00000 | $0.03000 |
| Opus 5.5 | $0.00000 | $0.01200 |
| Sonnet 5.5 | $0.00000 | $0.00600 |
| Haiku 4.5 | $0.00000 | $0.00300 |
Grade A, and why
build-screen scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 96 lines — stays where its author put it; the contents beside it link to each section on GitHub.
画面を作る
melta の契約から画面 1 枚を生成し、生成物を自分で lint して、「何を自動検査したか / 何を検査していないか」まで含めて報告する。依頼から提出までの往復を 1 回に潰すための手順書。
参照の実体は MCP ツールと AGENTS.md / DESIGN.md / design/contracts/ に置く。この手順書に仕様を複製しない(複製した瞬間に drift する)。
Step 1: 依頼の分類と契約の引き当て
AGENTS.mdの「## タスクベース読み込みガイド」の表を読み、依頼を該当行に当てる。表をここに写さない(AGENTS.mdが正)。複数の行に跨る依頼(例: 「サイドバー付きの設定画面」= サイドバー付きページ + 設定画面)は該当行の和集合を取る- 該当行が挙げるファイルを、表の順序どおりに読む
- MCP が使えるなら
get_component(契約の exact value)とsearch(どの契約が該当するかの探索)を優先する - MCP が無ければ
design/contracts/components/*.contract.jsonを直接読む - 値の正本は契約、原則の正本は
DESIGN.md。競合したら契約が勝つ(design/authority.md)
- MCP が使えるなら
- 引き当てた契約名と読んだファイルを控える。Step 5 でそのまま報告する
引き当てが 1 つも無い依頼(DS に契約が存在しないコンポーネントを含む)は、生成前にその旨を伝える。契約の無い部品を勝手に発明しない。
Step 2: 意図確認(最大 3 問)
下の質問バンクを上から見て、依頼文から答えが取れる問いは飛ばす。残ったものだけを 1 回でまとめて聞く。4 問以上聞かない。3 問を超える不確定さが残るなら、残りは仮置きして Step 5 の報告に「こう仮定した」と書く。
- Q1: 画面の主目的と主動作を 1 つ(例: 一覧から詳細へ / フォームを送信する)
- Q2: 含めるデータ状態(通常のみ / + 空状態 / + 読み込み中 / + エラー)
- Q3: 出力先パスと幅(ファイルパス / デスクトップのみ or レスポンシブ)
出力先が未指定なら提案する。examples/ は CI の Full Scan 対象なので、DS 公式サンプルにする意図が無い限り避ける。
拡張子は .html に揃える(Step 4 の CLI 経路が検査できるのは .html / .tsx / .jsx / .vue だけ。.htm などを渡すと検査自体が実行されない)。
ダークモードは聞かない。AGENTS.md の「テーマ・ダークモード」表の設定に従う。
Claude Code では AskUserQuestion で聞く。他のクライアントでは箇条書きで聞いて回答を待つ(推測で進めない)。
Step 3: 生成
DESIGN.mdの原則 + Step 1 で引き当てた契約の exact value で HTML を 1 枚書く。Tailwind class は契約の値を使う(近い値を目分量で選ばない)- Step 4 が拾えるルールも最初から守って書く(例: [TABLE_TH_SCOPE_REQUIRED] / [A11Y_NO_TABINDEX_POSITIVE] / [BTN_ICON_ONLY_ARIA_REQUIRED])。修正ループは検算であって、生成の手抜き分を回収する装置ではない
- severity が
errorでも detector がmanualのルールは Step 4 で絶対に捕まらない(例: 実効タップ標的 44px の下限 [A11Y_MIN_TAP_TARGET_44])。生成時に守るのが唯一の機会なので、使うコンポーネントのカテゴリのルールをdesign/contracts/rules.jsonで先に見る
Step 4: 自己検証(検査は最大 3 回 = 初回 + 修正後の再検査 2 回)
- 生成物を検査する。経路は 2 つあり、返ってくる情報が違う。どちらを使ったかを覚えておく(Step 5 の書き方が変わる)
- MCP 経路:
check_htmlに生成物を渡す。passed/violations/coverage.automated/coverage.notAutomatedが返る - CLI 経路(MCP が無い環境):
npm run design:lint-generated -- <生成ファイルのパス>。error があれば exit 1。返るのは違反一覧と件数と PASSED / FAILED だけで、passedフィールドも coverage も返らない
- MCP 経路:
- severity
errorを全部直して再検査する。warnは残してよいが Step 5 に列挙する - 3 回目の検査でも error が残るなら、残った violations を報告に載せて止まる。ルールを黙って緩めない・生成物を検査対象から外さない
- 検査そのものが実行できなかったとき(CLI が exit 2 = 対象拡張子でない / パスが解決できない / ruleset を読めない、ツールのエラー、MCP が応答しない)は、修正ループに入らない。生成物は無検査のままなので、エラー出力をそのまま持って Step 5 へ行く。「たぶん通る」で埋めない
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 96 lines · 0 tokens per session scan A 913c6277a218
build-screen is a skill published in the GitHub repository tsubotax/melta-ui (202 stars, last pushed 22d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 3,000 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-07.
Other skills, from other repositories
reimagine-it-lock
Design DNA capture — extracts palette, type stack, motifs, motion, and 3D signatures from a shipped HTML file and saves them as a reusable reference pack. Use when the user says /reimagine-it lock, "lock this design", "save design DNA", "capture this look", or wants to reuse a shipped design on a different target.…
ui-ux-design-pro
Senior-level UI/UX design expert for building data-driven, premium production interfaces. Use when you need to: 1. Design complex applications (dashboards, SaaS, AI tools) from scratch 2. Generate comprehensive design systems (tokens, palettes, typography) 3. Audit existing UI for quality, accessibility, and "craft"…
apple-design
Cross-platform UI/UX design reviewer grounded in Apple's Human Interface Guidelines (123 pages pulled from developer.apple.com, including 57 component pages) plus a design-craft lens for distinctive, non-templated work. Use it to audit, review, critique, or improve any mobile app (iOS, Flutter, React Native) or…
generate-figma-screen
Kod veya açıklamadan Figma'da tam ekran/sayfa oluşturur. Yayınlanmış design system bileşenlerini arayıp instance olarak yerleştirir; hardcode değer yerine DS token'larını kullanır. "Figma'da ekran oluştur", "kodu Figma'ya çevir", "landing page çiz", "ekran tasarla", "generate screen", "UI'ı Figma'ya aktar"…
fmcp-intent-router
F-MCP ile ilgili herhangi bir kullanıcı talebinin ilk giriş noktası. Kullanıcının niyetini analiz eder, hangi hedef SKILL'in çalıştırılacağına karar verir, o SKILL için gereken eksik input'ları tek turda toplar, özet+onay alır ve ondan sonra hedef SKILL'i çalıştırır. "figma", "ekran oluştur", "tasarım yap", "component…
figma-a11y-audit
Figma ekranını erişilebilirlik açısından denetler. Renk kontrastı (WCAG AA/AAA), minimum dokunma hedefi, fokus sırası, metin boyutu ve platform-bazlı ekran okuyucu önerileri (VoiceOver, TalkBack, ARIA) üretir. "a11y audit", "erişilebilirlik kontrol", "kontrast kontrol", "accessibility check", "ekran okuyucu spec"…